Vaner logoVaner

Model Context Protocol

Connect your AI client to Vaner.

Start with a uvx launcher snippet (no permanent install required), then configure the model backend and optional guardrails.

1. Connect your client (uvx-first)

Works without installing anything first. Use the picker, keep launcher on Run via uvx, paste the snippet, and restart your client.

No install required: uvx runs Vaner on demand inside your MCP client.

claude mcp add --transport stdio --scope user vaner -- uvx --from 'vaner[mcp]' vaner mcp --path .

Verify: Run `/mcp` inside a Claude Code session — `vaner` should appear with the five scenario tools.

Full guide: docs.vaner.ai/integrations/tools/claude-code-mcp

2. Pick a model backend

Local backends (Ollama, LM Studio, vLLM) run fully offline. Hosted backends need an API key in an environment variable.

Install Ollama and run `ollama pull llama3.2:3b`. Ollama exposes an OpenAI-compatible endpoint on port 11434.

[backend]
kind = "openai"
base_url = "http://127.0.0.1:11434/v1"
model = "llama3.2:3b"
api_key_env = ""

Drop the snippet into ~/.vaner/config.toml or run vaner init --backend-preset <id>.

3. Optional: bound ponder time

If you want Vaner to stop pondering after a fixed wall-clock budget, set a compute preset and an optional session cap.

Only when the machine is idle. Safe default.

[compute]
idle_only = true
cpu_fraction = 0.2
gpu_memory_fraction = 0.5
exploration_concurrency = 2
max_parallel_precompute = 1
max_cycle_seconds = 300

Prefer the CLI install?

If you want global CLI commands outside MCP, use the installer below.

curl -fsSL https://vaner.ai/install.sh | bash

Next